AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Continual Model Routing in Evolving Model Hubs

Jack Bell, Giacomo Carfi', Gerlando Gramaglia, Vincenzo Lomonaco

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Synthetic data and merging for CLIP · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add embedding anchoring (ablation)CARvE (10% domain replay) · CMRBench avgM-Acc / D-Fgt: 41.7 -> 43.1 M-Acc; 7.4 -> 5.9 D-FgtImproved
add projection anchoring (ablation)CARvE (10% domain replay) · CMRBench avgM-Acc / D-Fgt: 40.1 -> 43.1 M-Acc; 9.4 -> 5.9 D-FgtImproved
use domain-stratified coreset replay instead of random replay at 5% budget (ablation)CARvE · CMRBench avgM-Acc: 37.3 -> 40.5 M-Acc; 77.4 -> 78.5 D-AccImproved
use domain coreset replay instead of random replay at 10-20% budget (ablation)CARvE · CMRBench avgD-Acc: 80.6 -> 80.7 D-Acc at 10%; 82.8 -> 82.9 at 20%No clear effect
increase candidate set size K from 32 to 512 (ablation)CARvE · CMRBench avgD-Acc: 80.1 -> 81.0 D-AccNo clear effect
corrupt 10% of training labels (ablation)CARvE (10% domain replay) · CMRBench avgD-Acc: 80.7 -> 72.1 D-Acc; 5.9 -> 8.0 D-FgtHurt
corrupt 20% of training labels (ablation)CARvE (10% domain replay) · CMRBench avgD-Acc: 80.7 -> 63.1 D-AccHurt
replace random replay with CARvE at 10% replayLoRA LLaMA2-7B routing · CMRBench avgD-Acc / D-Fgt: 75.9 -> 80.7 D-Acc; 13.1 -> 5.9 D-FgtImproved
initialise model embeddings from model-card representations instead of randomly (ablation)CARvE · CMRBenchD-Acc: 3-5pp lower D-Acc, roughly double forgettingHurt
augment query with top-1 retrieved model card (RAT) instead of zero-shot (ablation)fine-tuned LLaMA2-7B router · CMRBenchaccuracyHurt
add EWC regularisation to CARvE (ablation)CARvE (10% replay) · CMRBench avgD-Acc: 80.7 -> 80.5No clear effect
use a smaller backbone (Qwen3-4B instead of LLaMA2-7B) (ablation)CARvE (10% replay) · CMRBench avgM-Acc: 43.1 -> 36.7 M-AccHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy